Abstract:
In the emerging trend, product developers and their customers use internet reviews as the primary tool for evaluating
products. Online communities, blogs, and public review websites provide a multitude of data about customers' overall
viewpoints, experiences, and opinions about goods. Product developers can harvest data on users' perceptions about their
preferred features and use that information to boost revenue and profit by planning and monitoring business strategies and
improving the overall quality of products. The reviews also assist prospective purchasers in making informed decisions on the
suitability of a product and pricing while reducing time and effort. Machine learning algorithms are used to identify and
categorize product evaluations. This paper presents an ensemble machine learning approach that integrates results drawn from
two base learners to improve accuracy in classification, which is the percentage of correctly classified product evaluation.
Multinomial Naïve Bayes and Unsupervised Lexicon were the base learners utilized to model the ensemble that was used to
classify consumer reviews as positive, neutral or negative. Feature extraction methods N-gram, Part of Speech, and features
from the lexical library TextBlob were used. The proposed model was evaluated on the real dataset for two items: the "Samsung
Galaxy A12" smartphone and the "Nissan Sentra" automobile brand and series. The experimental results indicate that the MNB
Lexicon Pooled Ensemble outperformed the individual MNB and Lexicon classifiers in rating prediction, with respective
accuracy, precision, recall and F1 measurements of 0.8250, 0.8932, 0.7970 and 0.8325.